Software Defined Compute Node (SDCN)
Publication Date: 2014-Feb-24
The IP.com Prior Art Database
This disclosure proposes a hardware component that forms a platform for a Software Defined Network Controller. While OpenFlow network controllers exist they require a standard server platform to host the openflow software. While this adds considerable function to the network it comes with significant structural limitations that impact its usefulness. This proposal is to build a server (compute node) on a module suitable for installation in a modular switch frame which leverages the switch back-plane. This positions the compute node in the fabric of the network with very low latency and high capacity. In addition to a standard server the module would be equipped with Field Programable Gate Arrays to provide special network acceleration functions to augment the capabilities of the OpenFlow controller.
Page 01 of 7
Software Defined Compute Node
Software Defined Compute Node
Software defined Networking or SDN is a recent approach to building network environments. The central principle to this approach is that networking and software applications are integrated as a whole rather than two largely distinct functions in use today. To improve both performance and flexibility a new class of SDN enabled network devices are being developed that can be manipulated by the software stack.
This in theory provides application performance benefits buy allowing the network to implement some of the software logic and /or the software directing through the network in more sophisticated ways than normally possible in the network infrastructure (switches, routers, etc).
In reality the current technology stack suffers from a huge disparity between the switching and forwarding speed of the network devices and the processing speed of the SDN (Openflow being an example) controllers. This is because Openflow relies on off the shelf server platforms designed as network edge devices as opposed to part of the network itself . The result is that the physical design and deployment of these servers wastes much of the potential performance gains in traffic direction by having poor connectivity to the very network fabric they are trying to direct. The illustration shows the common problem of wildly varying signal paths through the network mesh to switches under the control of a controller.
Page 02 of 7
Figure 1 - A single SDN controller deployment and variable signal paths.
The typical solution solution to problem path variability is to deploy multiple controllers.
Page 03 of 7
Figure 2 - A multiple Controller deployment
Multiple controllers however impact the ability of the system to have a single point of software control for the network limiting its
Page 04 of 7
value and causing complexity in the span of control for each device. Additionally the controllers rely on the very network they are
trying to operate to carry the signalling traffic to the SDN capable switches.
The Software Defined Network Compute Node
The design of large scale modular switches for data centres provides a way to address the current issues with SDN. These switches employ a very high speed back-plane which connects optional modules (sometimes described as blades) in a wide variety of possible configurations. This is done to allow upgrades and expansion of these switches over time and many network vendors use the same approach. The speed of data transport across the switch back-plane is orders of magnitude higher than the links used to connect adjacent switches in a network. For example, common links speeds use 10Gb/s (10x10^9) while back-plane speeds are often 10Tb/s (10x10^12).
This disclosure proposes specifically designed blade (called a Software Defined Network Compute Node) that can be inserted into a switch frame and connect...